An Entropy-Regularized ADMM For Binary Quadratic Programming

We propose an entropy regularized splitting model using low-rank factorization for solving binary quadratic programming with linear inequality constraints. Different from the semidefinite programming relaxation model, our model preserves the rank-one constraint and aims to find high quality rank-one...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of global optimization Jg. 87; H. 2-4; S. 447 - 479
Hauptverfasser: Liu, Haoming, Deng, Kangkang, Liu, Haoyang, Wen, Zaiwen
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York Springer US 01.11.2023
Springer
Schlagworte:
Algorithms
Computer Science
Mathematics
Mathematics and Statistics
Operations Research/Decision Theory
Optimization
Real Functions
Alternating direction method of multipliers
Entropy penalty
Binary quadratic programming
Riemannian manifold
Low-rank factorization
ISSN:0925-5001, 1573-2916
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:We propose an entropy regularized splitting model using low-rank factorization for solving binary quadratic programming with linear inequality constraints. Different from the semidefinite programming relaxation model, our model preserves the rank-one constraint and aims to find high quality rank-one solutions directly. The factorization transforms the variables into low-rank matrices, while the entropy term enforces the low-rank property of the splitting variable . A customized alternating direction method of multipliers is utilized to solve the proposed model. Specifically, our method uses the augmented Lagrangian function to deal with inequality constraints, and solves one subproblem on the oblique manifold by a regularized Newton method. Numerical results on the multiple-input multiple-output detection problem, the maxcut problem and the quadratic 0 - 1 problem indicate that our proposed algorithm has advantage over the SDP methods.
ISSN:0925-5001
1573-2916
DOI:10.1007/s10898-022-01144-0